Launch Day: 7 AI Agents Start Building Startups with $100 Each

Source: Dev.to

I just launched an experiment: 7 AI coding agents each receive $100 and 12 weeks to build a real startup from scratch—no human coding.

Agents Overview

Each agent autonomously:

1. Picks an idea.
2. Writes code.
3. Deploys the product.
4. Attempts to acquire users and generate revenue.

Key Findings

• Strategy beats code quality. Agents that planned distribution first outperformed those that wrote cleaner code.
• Product‑first planning: The Kimi agent drafted a full Product Hunt launch before writing any code.
• Simple stacks win. HTML + Tailwind deployments finished in hours, while Next.js agents spent days fixing build errors.
• Deploy loop is the bottleneck for AI agents.
• Context resets kill progress. Without persistent state, agents repeat mistakes. An orchestrator with structured state files mitigates this issue.

Orchestrator Details

A bash orchestrator manages the whole process:

• Cron‑scheduled 30‑minute sessions (2–8 per agent per day).
• Automatic git commits with [skip ci] on mid‑session commits.
• Deploy verification via health checks.
• Loop detection: same action three times → force an alternative.
• OpenRouter budget alerts sent to Discord.

All code is public on GitHub.

Monitoring & Reporting

• Live Dashboard – real‑time progress view.
• Daily Digest – hand‑written daily updates.
• Weekly Recaps – detailed analysis.

Additional Context

• The experiment was also launched on Product Hunt today.
• Full rules and additional documentation are available in the repository.

Which agent would you bet on?